一种基于条件相对平均熵的个性化推荐算法

为了提高现有推荐算法的准确性,提出一种基于条件相对平均熵的个性化推荐算法.首先,采用加权的CNM算法构建复杂网络,挖掘该网络的社团结构,作为商品待推荐域;其次,利用条件互信息和条件相对平均熵生成有效的节点次序,以提升贝叶斯网络构建的准确性;然后采用K2算法学习贝叶斯网络,分析出用户的消费性格,并判断待推荐域中商品与消费性格的联系,向用户提供感兴趣和合理的推荐;最后,基于电信运营商的实际数据进行实证分析,验证了该算法的有效性.

A personalized recommendation algorithm based on conditional relative average entropy

In order to improve the accuracy of recommendation algorithm,one personalized recommendation algorithm based on conditional relative average entropy is presented.First of all,through weighted CNM algorithm we construct complex network and excavate the network’s community structure. The result is regarded as the uncertain recommendation domain. Further more,conditional mutual information and conditional relative average entropy are used to determine the effective node ordering as input of K2 algorithm,which can improve the accuracy of Bayesian network construction,and then learn Bayesian network by K2 algorithm and analyze the consumer characteristics. We use the relationship between the commodity and the consumer characteristics to confirm the recommendation domain.Finally,the empirical analysis of the actual data of the telecom operators is carried out to verify the validity of the above algorithm.